2026 State of AI Report: The Builder's Economy
How 300+ Executives Are Scaling AI Products
A year ago, boards wanted to hear the AI strategy. Now they want AI unit economics.
That shift, from whether AI works to whether it pays, is the through-line of this report. AI is no longer a differentiator on its own. It has become a core expectation. The challenge has moved from launching AI features to building businesses that can sustain and scale them.
Our findings draw on surveys of over 300 executives at software companies building AI products, including CEOs, heads of engineering, heads of AI, and heads of product. We also include perspectives from the ICONIQ community.
The data is clear to us: builders have converged on the application layer, agentic capabilities are now a top investment priority, and the economics are following. Margins are expanding, pricing models are being rebuilt to reflect actual usage and outcomes, and the gap between high-growth companies and everyone else appears to be widening.
Welcome to the Builder’s Economy.
Where are AI companies building in 2026?
The center of gravity has moved towards to the application layer. Close to two-thirds of the builders we surveyed are working on horizontal or vertical AI applications, with vertical alone at 43%, and momentum moving toward complex, regulated domains. Products targeting financial services and healthcare use cases rose the fastest, because a product embedded in a real workflow appears to be harder to dislodge than a general-purpose model sitting one layer up.
Underneath those products, multi-model is now generally the default. Companies are running an average of about 3.3 models, with licensed 3rd-party APIs as the most used model type.
Nearly half choose to use 2+ model types, suggesting multi-model strategies are becoming the default. Anthropic became the most-cited provider, climbing from 51% to 81% adoption in six months.
That enterprise pressure is exposing a gap in trust infrastructure. Quality assurance is still mostly reactive, caught through user feedback at 67% and model monitoring at 63%, rather than with testing that proactively identifies problems—proactive adversarial testing remains a minority practice at 21%. Data-protection guardrails are nearly universal, but defenses against AI-specific risks lag, with prompt injection detection at 44% and data exfiltration prevention at 38%.
.png)
Are AI products paying off?
The clearest evidence that AI is creating value is that it shows up in the financials. AI products grew from 32% of revenue in 2025 to a projected 42% this year, and are on track for roughly 53% by 2027. Gross margins are improving, jumping from 45% in 2025 to a projected 53% in 2026, and 59% in 2027.
Pricing is being rebuilt to match. Subscriptions remain most common, but consumption-based pricing rose from 35% to 42% in six months, and outcome-based from 18% to 23%. On average, companies are blending 1.7 pricing models. On the cost side, two-thirds report improved per-query unit economics, helped by managing inference costs, improved model routing strategies, and revenue growth that creates cost leverage. As products scale, talent's share of cost falls while inference rises.
How is AI reshaping the org chart in 2026?
AI is changing how companies are built as much as what they build. Seventy-eight percent of companies are rethinking workforce planning, with 45% expecting a different mix of roles and 33% expecting smaller teams. Companies that earn most of their revenue from AI run flatter, with 72% reporting four or fewer management layers between the CEO and the most junior individual contributor, versus 56% of other peers. Read more about key themes centered around how AI companies are redesigning their organizations in our “From Org Charts to Outcomes: The New Operating Model for Talent” article.
Growth is uneven by function, with R&D, sales, and product and design expanding, while customer support and G&A have contracted. Hiring follows suit: AI product managers and solutions engineers are now mainstream, while forward-deployed engineers and AI safety and reliability roles are scaling fastest. The forward-deployed engineer has graduated from a services patch to a permanent go-to-market motion, with half of companies planning to scale it and most treating it as a revenue role tied to expansion and retention.
This has become a live operating challenge rather than a planning question. The debate is less about how many engineers and more about which roles still create value. An engineer who turns a design doc into a feature is the most exposed, since AI already does much of that work well, while those who specialize in infrastructure and security, or operate as a product-engineering hybrid who can ship, have held their value.
How much do companies spend on internal AI, and is it working?
The companies that appear to be pulling ahead have integrated aI into workflows for compounding benefits.
Spending on AI for internal productivity, including direct and indirect spend, is projected to rise from 11% of revenue in 2025 to 16% in 2026, heading toward 19% in 2027. And many companies underestimate the true cost.
One of the biggest surprises came from token spend in agentic pipelines, where one builder watched a workflow budgeted at ten cents a run drift past a dollar fifty as agents retried and corrected themselves.
This is followed by data infrastructure and enablement. The high-growth advantage then shows up throughout the process: top performers stand up a new AI tool in about 2.5 months versus 3.5, write 59% of their code with AI versus 47%, and see 48% productivity gains from coding assistance vs 32%. and report a 16-point larger productivity gain from coding assistance, where their peers reach 48%. Each edge makes the next one easier to win.
Measuring token counts was also named as a poor success metric, since it distorts behavior. Charging usage back to cost centers works better, and traditional measures like cycle time and deploys per week still hold up as velocity proxies. One early-stage company runs the aggressive version, offering unlimited tokens, but managers owe three to five times the productivity and own delivery accountability. Top engineers there run up to $8,000 a day. Read more on how to measure success in the Age of AI in the “AI Adoption Index” template.
Agents are a clear test of how far internal AI has come. They have spread quickly, especially among larger companies, but average productivity gains sit below 30% across every revenue band. The gap is reliability. Almost half of companies say their agents still need a human to step in on at least 30% of tasks, and the most common failure is multi-step workflows that break partway through. The technology is real, but closing the distance between deploying an agent and trusting it unattended is what the next stage of development is about.
The companies pulling ahead in our survey have made pricing, cost structure, and organizational design into compounding advantages. The data shows how.
Download the full 2026 State of AI report for the complete picture.